The equilibrium nonexistence problem in Rothschild and Stiglitz's insurance market is reexamined in a dynamic setting. Insurance firms are boundedly rational and offer menus of insurance contracts which are periodically revised: profitable competitors' contracts are imitated and loss-making contracts are withdrawn. Occasionally, a firm experiments by withdrawing or innovating a random set of contracts. We show that Rothschild and Stiglitz's candidate competitive equilibrium contracts constitute the unique long-run market outcome if innovation experiments are restricted to contracts which are sufficiently "similar" to those currently on the market. (C) 2002 Elsevier Science (USA). All rights reserved.